Content extraction with apache tika

Content Extraction with Apache Tika
Jukka Zitting | Tika committer, co-author of Tika in Action

© 2012 Adobe Systems Incorporated. All Rights Reserved.

Content Extraction with Apache Tika

 Introduction to Apache Tika
 Full text extraction with Tika
 Tika and Solr - the ExtractingRequestHandler
 Tika and Lucene - direct feeding of the index
 forked parsing
 link extraction

© 2012 Adobe Systems Incorporated. All Rights Reserved. 2

Introduction to Apache Tika
section 1 / 4


Introduction to Apache Tika

The Apache Tika™ toolkit
- detects and extracts
- metadata and structured text content
- from various documents
- using existing parser libraries.


Problem domain


The Tika solution

It is a truth
universally
acknowledged, that
a single man in
possession of a
good fortune, must
be in want of a
wife...
Content
dc:title=
Pride and Prejudice
dc:creator=
Jane Austen
Document dc:date=1813

Metadata


Project background

 Brief history
 2007 Tika started in the Apache Incubator
 2008 Tika graduates into a Lucene subproject
 2010 Tika becomes a standalone TLP
 2011 Tika 1.0 released
 2011 Tika in Action published
 Latest release is Apache Tika 1.2
 thousands of known media types
 most with associated type detection patterns
 dozens of supported document formats
 including all major office formats
 basic language detection
 etc.
 For more information http://tika.apache.org/


Ohloh summary (http://www.ohloh.net/p/tika)


Full text extraction with Tika
section 2 / 4


Demo: tika-app-1.2.jar

 https://github.com/jukka/tika-demo
 java -jar tika-app-1.2.jar


tika-app as a command line tool

$ java -jar tika-app-1.2.jar --xhtml /path/to/
document.doc

$ java -jar tika-app-1.2.jar --text http://example.com/
document

$ java -jar tika-app-1.2.jar --metadata < document.doc

$ cat document.doc | java -jar tika-app-1.2.jar --text |
grep foo

$ java -jar tika-app-1.2.jar --help


Tika’s Java API

 Divided in two layers
 The Tika facade: org.apache.tika.Tika
 Lower-level interfaces like Parser, Detector, etc.

 Use the Tika facade by default
 Provides simple support for most common use cases
 Example: new Tika().parseToString(“/path/to/document.doc”)

 Use the lower-level interfaces for more power or ﬂexibility
 Allows ﬁne-grained control of Tika functionality
 More complicated programming model
 Parsed content handled as XHTML SAX events
 Not all functionality is exposed through the Tika facade


Tika and Solr - the ExtractingRequestHandler
section 3 / 4


ExtractingRequestHandler

 aka Solr Cell
 http://wiki.apache.org/solr/ExtractingRequestHandler

“Solr's ExtractingRequestHandler uses Tika
to allow users to upload binary files to Solr and
have Solr extract text from it and then index it.”

 For example:
$ curl "http://localhost:8983/solr/update/extract?
literal.id=document&commit=true" -F "file=@document.doc"

 Supports both text and metadata extraction
 with plenty of configurable options


ExtractingRequestHandler parameters

 Helping Tika do it’s job
 resource.name=document.doc - Helps Tika’s automatic type detection
 resource.password=secret - Allows Tika to read encrypted documents
 passwordsFile=/path/to/password-file - Resource name to password
mappings
 for example: .*.pdf$ = pdf-secret
 Capturing special content
 xpath=//a - Capture only content inside elements that match the
specified query
 capture=h1 - Capture content inside specific elements to a separate
field
 captureAttr=true - Capture attributes into separate fields named after
the element
 Mapping field names
 lowernames=true - Normalize metadata field names to “content_type”,
etc.


Tika and Lucene - direct feeding of the index
section 4 / 4


Using the Tika facade to feed Lucene

// Index ﬁrst part of the document
String text = new Tika().parseToString(“/path/to/document.doc”);
document.add(new TextField(“text”, text, Field.Store.NO));

// Index the full document
Reader reader = new Tika().parse(“/path/to/document.doc”);
document.add(new TextField(“text”, reader));

// Index also some metadata
Metadata metadata = new Metadata();
Reader reader = new Tika().parse(
new FileInputStream(“/path/to/document.doc”), metadata);
document.add(new TextField(“text”, reader));
document.add(new StringField(“type”,
metadata.get(Metadata.CONTENT_TYPE));


Things to consider

 What if the document is larger than your memory?
 Index only ﬁrst N bytes/characters?
 WriteOutContentHandler supports an explicit write limit
 Enabled by default in the Tika facade, see get/setMaxStringLength()

 What if the document is malformed or intentionally broken?
 Could cause denial of service problems
 Might even crash the entire JVM due to bugs in native libraries in the JDK!
 SecureContentHandler monitors parsing and terminates it if things look
bad
 Enabled by default in the Tika facade

 Ultimate solution: forked parsing and the Tika server
 Parse documents in separate, sandboxed JVM processes
 A document could fail to parse, but your application won’t crash
 Code is already there, but still a bit tricky to set up


Link extraction for web crawlers

 The LinkContentHandler class can be used to extract all links from
a document
 Works also with links in things like PDF, MS Word and email documents
 Use TeeContentHandler to combine with other ways of capturing content

// for example
LinkContentHandler lch = new LinkContentHandler();
BodyContentHandler bch = new BodyContentHandler();
new Tika().getParser().parse(..., new TeeContentHandler(lch, bch), ...);

System.out.println(“Content: “ + bch);
for (Link link : lch.getLinks()) {
System.out.println(“Link: “ + link):
}


Questions?
http://tika.apache.org/


Content extraction with apache tika

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (20)

Similaire à Content extraction with apache tika

Similaire à Content extraction with apache tika (20)

Plus de Jukka Zitting

Plus de Jukka Zitting (19)

Content extraction with apache tika

Notes de l'éditeur